[Bf-blender-cvs] SVN commit: /data/svn/bf-blender [27125] branches/render25/source/blender/ render/intern/source/cache.c: Render Cache: irradiance cache:
Brecht Van Lommel
brecht at blender.org
Wed Feb 24 18:27:29 CET 2010
Revision: 27125
http://projects.blender.org/plugins/scmsvn/viewcvs.php?view=rev&root=bf-blender&revision=27125
Author: blendix
Date: 2010-02-24 18:27:29 +0100 (Wed, 24 Feb 2010)
Log Message:
-----------
Render Cache: irradiance cache:
* increase truncated SVD epsilon to avoid avoid large color changes when
interpolating between near colinear points.
* added neighbour clamping. it doesn't seem to help much and make things
a bit slower at the moment, but hopefully can be used to get better
distribution later.
Main remaining causes of artifacts seem to be:
* even 32x32 (1024) samples can still be noisy leading to low frequency
noise after interpolation
* borders between tiles can have discontinuities. this is mostly visible
in combination with the above problem, i.e. if the result is splotchy
it will have a different splotches in the next tile.
* very small features may be missed still, leading to missing shadows in
some cases or even stippled patterns instead if they are detected part
of the time, neighbour clamping helps somewhat here.
Modified Paths:
--------------
branches/render25/source/blender/render/intern/source/cache.c
Modified: branches/render25/source/blender/render/intern/source/cache.c
===================================================================
--- branches/render25/source/blender/render/intern/source/cache.c 2010-02-24 17:14:16 UTC (rev 27124)
+++ branches/render25/source/blender/render/intern/source/cache.c 2010-02-24 17:27:29 UTC (rev 27125)
@@ -407,8 +407,9 @@
#define MAX_PIXEL_DIST 10.0f
#define MAX_ERROR_K 1.0f
#define WEIGHT_NORMAL_DENOM 65.823047821929777f /* 1/(1 - cos(10°)) */
-#define HORIZON_CUTOFF 0.17364817766693041f /* cos(80°) */
+#define SINGULAR_VALUE_EPSILON 1e-4f
#define LSQ_RECONSTRUCTION
+#define NEIGHBOUR_CLAMP
/* Data Structures */
@@ -471,7 +472,7 @@
void TNT_svd(float m[][4], float *w, float u[][4]);
-int svd_invert_m4_m4(float R[4][4], float M[4][4])
+void svd_invert_m4_m4(float R[4][4], float M[4][4])
{
float V[4][4], W[4], Wm[4][4], U[4][4];
@@ -479,16 +480,14 @@
TNT_svd(V, W, U);
zero_m4(Wm);
- Wm[0][0]= (W[0] < 1e-6f)? 0.0f: 1.0f/W[0];
- Wm[1][1]= (W[1] < 1e-6f)? 0.0f: 1.0f/W[1];
- Wm[2][2]= (W[2] < 1e-6f)? 0.0f: 1.0f/W[2];
- Wm[3][3]= (W[3] < 1e-6f)? 0.0f: 1.0f/W[3];
+ Wm[0][0]= (W[0] < SINGULAR_VALUE_EPSILON)? 0.0f: 1.0f/W[0];
+ Wm[1][1]= (W[1] < SINGULAR_VALUE_EPSILON)? 0.0f: 1.0f/W[1];
+ Wm[2][2]= (W[2] < SINGULAR_VALUE_EPSILON)? 0.0f: 1.0f/W[2];
+ Wm[3][3]= (W[3] < SINGULAR_VALUE_EPSILON)? 0.0f: 1.0f/W[3];
transpose_m4(V);
mul_serie_m4(R, U, Wm, V, 0, 0, 0, 0, 0);
-
- return 1;
}
void lsq_4D_solve(Lsq4DFit *lsq, float solution[CACHE_DIMENSION])
@@ -576,6 +575,48 @@
if(indirect) copy_v3_v3(indirect, C+4);
}
+#ifdef NEIGHBOUR_CLAMP
+/* Neighbour Clamping [Křivánek 2006] */
+
+static void irr_cache_clamp_sample(IrrCache *cache, IrrCacheSample *nsample)
+{
+ IrrCacheSample *sample;
+ IrrCacheNode *node, **stack, **stacknode;
+ float P[3];
+ int i;
+
+ copy_v3_v3(P, nsample->P);
+
+ stack= cache->stack;
+ stacknode= stack;
+
+ *stacknode++= &cache->root;
+ while(stacknode > stack) {
+ node= *(--stacknode);
+
+ /* sum the values in this level */
+ for(sample=node->samples; sample; sample=sample->next) {
+ float l;
+
+ /* avoid issues with coincident points */
+ l= len_squared_v3v3(sample->P, P);
+ l= (l > EPSILON)? sqrtf(l): EPSILON;
+
+ nsample->dP= minf(nsample->dP, sample->dP + l);
+ sample->dP= minf(sample->dP, nsample->dP + l);
+ }
+
+ /* check the children */
+ for(i=0; i<8; i++) {
+ IrrCacheNode *tnode= node->children[i];
+
+ if(tnode && irr_cache_node_point_inside(tnode, 4.0f, 0.0f, P))
+ *stacknode++= tnode;
+ }
+ }
+}
+#endif
+
/* Add a Sample */
static void irr_cache_add(Render *re, ShadeInput *shi, IrrCache *cache, float *ao, float env[3], float indirect[3], float P[3], float dPdu[3], float dPdv[3], float N[3])
@@ -601,6 +642,12 @@
sample->dP= Rmean;
irr_sample_set(sample->C, ao, env, indirect);
+#ifdef NEIGHBOUR_CLAMP
+ /* neighbour clamping trick */
+ irr_cache_clamp_sample(cache, sample);
+ Rmean= sample->dP; /* copy dP back so we get the right place in the octree */
+#endif
+
/* error multiplier */
Rmean /= MAX_ERROR_K;
@@ -642,18 +689,21 @@
/* Lookup */
+#include <fenv.h>
+
int irr_cache_lookup(Render *re, ShadeInput *shi, IrrCache *cache, float *ao, float env[3], float indirect[3], float cP[3], float dPdu[3], float dPdv[3], float cN[3], int preprocess)
{
IrrCacheSample *sample;
IrrCacheNode *node, **stack, **stacknode;
float accum[CACHE_DIMENSION], P[3], N[3], totw;
- float discard_weight, maxdist;
+ float discard_weight, maxdist, distfac;
int i, added= 0, totfound= 0;
#ifdef LSQ_RECONSTRUCTION
Lsq4DFit lsq;
#endif
/* XXX check how often this is called! */
+ /* XXX can identical samples end up in the cache now? */
/* a small value for discard-smoothing of irradiance */
discard_weight= (preprocess)? 0.1f: 0.0f;
@@ -676,9 +726,16 @@
memset(accum, 0, sizeof(accum));
#endif
- maxdist= sqrtf(len_v3(dPdu)*len_v3(dPdv)); //*0.5f);
- maxdist *= (preprocess? MAX_PIXEL_DIST/2: MAX_PIXEL_DIST);
+ /* the motivation for this factor is that in preprocess we only require
+ one sample for lookup not to fail, whereas for least squares
+ reconstruction we need more samples for a proper reconstruction. it's
+ quite arbitrary though and it would be good to have an actual
+ guarantee that we have enough samples for reconstruction */
+ distfac= (preprocess)? 1.0f: 2.0f;
+ maxdist= sqrtf(len_v3(dPdu)*len_v3(dPdv))*0.5f;
+ maxdist *= MAX_PIXEL_DIST*distfac;
+
while(stacknode > stack) {
node= *(--stacknode);
@@ -696,10 +753,8 @@
dist= len_v3(D);
- if(preprocess && !(dist < sample->dP))
+ if(dist > sample->dP*distfac)
continue;
- else if(!preprocess && !(dist < sample->dP*2))
- continue;
#if 0
{
@@ -720,6 +775,7 @@
/* compute the weight */
w= 1.0f - MAX_ERROR_K*maxf(e1, e2);
+
if(w > BLI_thread_frand(cache->thread)*discard_weight) {
if(!preprocess) {
#ifdef LSQ_RECONSTRUCTION
@@ -783,7 +839,7 @@
{
RenderResult *rr= pa->result;
IrrCache *cache;
- int offs, crop, x, y, seed;
+ int crop, x, y, seed, step;
if(!((re->db.wrld.aomode & WO_AOCACHE) && (re->db.wrld.mode & (WO_AMB_OCC|WO_ENV_LIGHT|WO_INDIRECT_LIGHT))))
return;
@@ -793,62 +849,62 @@
seed= pa->rectx*pa->disprect.ymin;
- offs= 0;
crop= 0;
- if(pa->crop) {
+ if(pa->crop)
crop= 1;
- offs= pa->rectx + 1;
- }
- rr->renrect.ymin= 0;
- rr->renrect.ymax= -2*crop;
- rr->renlay= rl;
+ //step= MAX2(pa->disprect.ymax - pa->disprect.ymin + 2*crop, pa->disprect.xmax - pa->disprect.xmin + 2*crop);
+ step= 1;
- for(y=pa->disprect.ymin+crop; y<pa->disprect.ymax-crop; y++, rr->renrect.ymax++) {
- for(x=pa->disprect.xmin+crop; x<pa->disprect.xmax-crop; x++) {
- int lx = (x - pa->disprect.xmin);
- int ly = (y - pa->disprect.ymin);
- int od = lx + ly*(pa->disprect.xmax - pa->disprect.xmin);
- PixelRow row[MAX_PIXEL_ROW];
- int a, b, totrow;
+ while(step > 0) {
+ rr->renrect.ymin= 0;
+ rr->renrect.ymax= -2*crop;
+ rr->renlay= rl;
- BLI_thread_srandom(pa->thread, seed++);
+ for(y=pa->disprect.ymin+crop; y<pa->disprect.ymax-crop; y+=step, rr->renrect.ymax++) {
+ for(x=pa->disprect.xmin+crop; x<pa->disprect.xmax-crop; x+=step) {
+ int lx = (x - pa->disprect.xmin);
+ int ly = (y - pa->disprect.ymin);
+ int od = lx + ly*(pa->disprect.xmax - pa->disprect.xmin);
+ PixelRow row[MAX_PIXEL_ROW];
+ int a, b, totrow;
- /* create shade pixel row, sorted front to back */
- totrow= pixel_row_fill(row, re, pa, od);
+ BLI_thread_srandom(pa->thread, seed++);
- for(a=0; a<totrow; a++) {
- shade_samples_from_pixel(re, ssamp, &row[a], lx+pa->disprect.xmin, ly+pa->disprect.ymin);
+ /* create shade pixel row, sorted front to back */
+ totrow= pixel_row_fill(row, re, pa, od);
- for(b=0; b<ssamp->tot; b++) {
- ShadeInput *shi= &ssamp->shi[b];
- ShadeGeometry *geom= &shi->geometry;
- float *ao= (re->db.wrld.mode & WO_AMB_OCC)? shi->shading.ao: NULL;
- float *env= (re->db.wrld.mode & WO_ENV_LIGHT)? shi->shading.env: NULL;
- float *indirect= (re->db.wrld.mode & WO_INDIRECT_LIGHT)? shi->shading.indirect: NULL;
- int added;
+ for(a=0; a<totrow; a++) {
+ shade_samples_from_pixel(re, ssamp, &row[a], lx+pa->disprect.xmin, ly+pa->disprect.ymin);
- added= irr_cache_lookup(re, shi, cache,
- ao, env, indirect,
- geom->co, geom->dxco, geom->dyco, geom->vn, 1);
-
- if(added) {
- if(indirect)
- copy_v3_v3(rl->rectf + od*4, indirect);
- else if(env)
- copy_v3_v3(rl->rectf + od*4, env);
- else if(ao)
- rl->rectf[od*4]= *ao;
+ for(b=0; b<ssamp->tot; b++) {
+ ShadeInput *shi= &ssamp->shi[b];
+ ShadeGeometry *geom= &shi->geometry;
+ float *ao= (re->db.wrld.mode & WO_AMB_OCC)? shi->shading.ao: NULL;
+ float *env= (re->db.wrld.mode & WO_ENV_LIGHT)? shi->shading.env: NULL;
+ float *indirect= (re->db.wrld.mode & WO_INDIRECT_LIGHT)? shi->shading.indirect: NULL;
+ int added;
+
+ added= irr_cache_lookup(re, shi, cache,
+ ao, env, indirect,
+ geom->co, geom->dxco, geom->dyco, geom->vn, 1);
+
+ if(added) {
+ if(indirect)
+ add_v3_v3(rl->rectf + od*4, indirect);
+ if(env)
+ add_v3_v3(rl->rectf + od*4, env);
+ if(ao)
+ rl->rectf[od*4] += *ao;
+ }
}
}
}
- offs++;
+ if(re->cb.test_break(re->cb.tbh)) break;
}
- offs+= 2*crop;
-
- if(re->cb.test_break(re->cb.tbh)) break;
+ step /= 2;
}
memset(rl->rectf, 0, sizeof(float)*4*rl->rectx*rl->recty);
More information about the Bf-blender-cvs
mailing list